EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

The effect of horizontal diffusion parameterization in convection-permitting REMO-NH simulations over Germany

Thomas Frisius, Daniela Jacob, Armelle Reca Remedio, Kevin Sieck, and Claas Teichmann
Thomas Frisius et al.
  • Helmholtz-Zentrum Geesthacht, Climate Service Center Germany (GERICS) , Germany (

Moving towards convection permitting simulations up to few kilometers scale are emerging solutions to the challenge and complexities in simulating different convective phenomena especially over mountainous regions. In this study we execute sensitivity experiments with the non-hydrostatic regional climate model REMO-NH at convection permitting resolution (~3km). We use this model in three setups where different parameterization schemes for horizontal diffusion are tested. In the first setup “DIFF2” we utilize the standard 2nd order diffusion while the second setup “DIFF4” applies 4th order diffusion. The higher order has a smaller impact on larger scales so that the atmospheric fields exhibit more details, especially in regions with high convective activity. In the third setup “TURB3D”, REMO-NH runs with a new 3D Smagorinsky-type turbulence scheme instead of the artificial diffusion schemes. Though turbulent horizontal diffusion is of second order in this setup, it incorporates a spatially and temporally varying exchange coefficient so that flows with little deformation remain unaffected. The domain of the simulations driven with EURO-CORDEX boundary data covers Germany and the time integration spans the year 2006.

Selected cases reveal a better representation of convective elements in DIFF4 and TURB3D when compared with DIFF2. We cannot compare these individual cases directly to observations since REMO-NH is not a reanalysis but a climate model. However, the spatial precipitation fields deduced from DWD radar data have characteristics which are more similar to DIFF4 and TURB3D than to DIFF2. More details are resolved in DIFF4 and TURB3D since the diffusion mainly act at the smallest spatial scales resolved by the model. DIFF2 smoothes convective activity drastically so that it appears in the form of unrealistically wide convective cells. On the other hand, the statistics of precipitation (seasonal average, standard deviation and 95th percentile) show a better agreement with observations in the simulation DIFF2 and TURB3D. TURB3D appears to be the best compromise regarding the simulation of precipitations fields. However, TURB3D exhibits a warm bias in the 2m temperature field in autumn and winter. Further model development may help to overcome this issue.

How to cite: Frisius, T., Jacob, D., Reca Remedio, A., Sieck, K., and Teichmann, C.: The effect of horizontal diffusion parameterization in convection-permitting REMO-NH simulations over Germany, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-17888,, 2020

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 1 – uploaded on 07 May 2020
  • CC1: Increased T2M when using Turb3D, Marvin Kähnert, 08 May 2020

    Hey Thomas,

    very interesting work! To pick up the point from the group chat again: In the absence of shallow convection we see an increase in near-surface temperatures in our model runs (AROME-Arctic) due to the reduction in vertical mixing in convective PBLs (efficiency of non-local transport vs local one). In general the atmosphere tends to be more unstable.

    Do you observe a similar behavoir for your Turb3D runs compared to pendants with active shallow convection? Or do you relate the increased T2M in autumn/winter to stable PBL features (which would imply too strong of a mixing by the scheme)?

    • AC1: Reply to CC1, Thomas Frisius, 08 May 2020

      Hello Marvin,

      thank you for your interest in out work. In the experiments no shallow convection scheme was used. Indeed, the second possibility that increased T2m results from too strong mixing could deliver an explanation for the bias in TURB3D. I will investigate this issue by varying the parameters of the scheme.

      Best regards,